2408 IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 51, NO. 10, OCTOBER 2016

A Full-Duplex Single-Chip Transceiver WithSelf-Interference Cancellation in 0.13 μmSiGe BiCMOS for Electron Paramagnetic

Resonance SpectroscopyXuebei Yang and Aydin Babakhani

Abstract— This paper presents a miniaturized EPR spectrom-eter based on a single-chip transceiver. Utilizing a novel on-chipself-interference cancellation circuit, the electromagnetic couplingfrom the transmitter (TX) to the receiver (RX) is minimized,allowing simultaneous achievement of large TX output power andlow RX noise figure (NF). In the measurement, the RX achievesa NF of 3.1 dB/6.3 dB at 10 MHz/50 kHz baseband frequencies,when the TX and cancellation circuits are turned off. Themeasured flicker noise corner is 60 kHz, more than 10× lowerthan the prior work. Moreover, for the first time, the operationof the RX and cancellation circuit is demonstrated when a co-integrated TX is operating at the same time and frequency, whileproducing >20 dBm output power. When the TX and cancellationcircuits are turned on, at −10 dBm interference power, themeasured NF is 6.8 dB/11.1 dB at 10 MHz/50 kHz basebandfrequencies. This is lower by 5.6 dB/9.6 dB at 10 MHz/50 kHzbaseband frequencies, compared to the NF with the cancellationcircuit off at the same interference power. The transceiver chip isimplemented in IBM 0.13 µm BiCMOS process and consumes apower of 2 W. Utilizing this transceiver, an electron paramagneticresonance (EPR) spectrometer is built and tested. It is observed,through measurement, that the interference cancellation circuitincreases the signal-to-noise ratio (SNR) of the EPR signal by7 dB at −10 dBm interference power. Compared to prior work,the reported EPR spectrometer improves the sensitivity of thesystem by 25 dB.

Index Terms— BiCMOS, electron paramagnetic resonance,electron spin resonance, full-duplex, self-interference cancella-tion, SiGe, silicon, transceiver.

I. INTRODUCTION

THE electron paramagnetic resonance (EPR) phenomenonis based on the interaction of an electron magnetic

dipole with an electromagnetic field, in the presence of a DCmagnetic field. Since the first observation in the mid-1940s [1],EPR spectroscopy has found its application in numerous fields.In clinical medicine, EPR spectroscopy is the only direct andnon-invasive method that can monitor the partial oxygen pres-sure (pO2) in human vessels and tissues. This measurementprovides vital information in the treatment of cardiovasculardiseases and during the radiation therapy [2]–[4]. In the oil

Manuscript received March 20, 2016; revised May 17, 2016 andJuly 6, 2016; accepted July 8, 2016. Date of publication August 18, 2016;date of current version September 30, 2016. This paper was approved byAssociate Editor Pietro Andreani.

The authors are with Rice University, Houston, TX 77005 USA.Color versions of one or more of the figures in this paper are available

online at http://ieeexplore.ieee.org.Digital Object Identifier 10.1109/JSSC.2016.2592998

and gas industry, EPR spectroscopy is used to determine theconcentration of various paramagnetic chemicals in crude oil,such as asphaltenes, corrosion ions, and vanadyl complexesVO2+ [5], [6]. In the semiconductor industry, EPR spec-troscopy provides a powerful tool for detecting point defects insemiconductor lattices, including silicon, silicon carbide, zincoxide [7], [8].

Unfortunately, while EPR spectroscopy has demonstratedits significance in various fields, this technique has onlybeen exploited in few research laboratories. This is causedprimarily by the deficiencies of existing commercial EPRspectrometers. Traditionally, an EPR spectrometer consistsof a large number of discrete electrical components, whichare usually built with dedicated and expensive processes.As a result, the cost, weight, and size of traditional EPRspectrometers are prohibitively large, restricting their use inhand-held applications [22].

In order to reduce the cost, weight, and size of EPRspectrometers, recent efforts have focused on integrating theelectrical components of the EPR spectrometer through the useof silicon integrated-circuit technology [9]–[11]. For example,in [11] the RF transceiver of an EPR spectrometer wasintegrated onto a single chip, which represents the highestlevel of integration so far. However, the EPR spectrometerreported in [11] suffered from relatively low sensitivity, mostlydue to a TX self-interference problem. In order to mitigatethe TX self-interference problem, in [11] the TX power wasdeliberately lowered to about 5 dBm. Unfortunately, existingstudies have shown that the EPR signal power generatedby the sample is proportional to the TX power sent to thesample [12]. Therefore, the EPR spectrometer reported in [11]has a relatively poor sensitivity.

In this paper, we propose a technique to enhance thesensitivity of the EPR spectrometer by mitigating the TX self-interference. Although TX self-interference is a rather generalproblem that would appear in all in-band full-duplex appli-cations, the unique operating condition of EPR spectroscopyposes specific challenges to the transceiver. In EPR spec-troscopy, the TX interference is an unmodulated continuous-wave signal, which to some extent, relaxes the linearity andintermodulation requirement for the RX. However, the fre-quency of the baseband signal is typically lower than 100 kHz,which poses stringent requirement to the RX noise figure (NF)

YANG AND BABAKHANI: FULL-DUPLEX SINGLE-CHIP TRANSCEIVER WITH SELF-INTERFERENCE CANCELLATION IN 0.13 μm SiGe BiCMOS 2409

at low baseband frequencies. In prior in-band full-duplextransceivers [13]–[17] that targeted wireless communicationapplications, mixer-first or current-mode mixer-first architec-tures were mostly utilized. These architectures unfortunatelyresulted in a poor RX NF especially at low baseband fre-quencies. As a result, TX interference-resilient techniquesbased on mixer-first architectures cannot be directly applied inEPR spectroscopy. Furthermore, in prior interference-resilient-transceiver papers, the TX is either not implemented on thesame chip, or is integrated on the same chip but remains offduring RX measurements. Therefore, the true performanceof the RX remains unknown when a co-integrated TX ison and operating at the same frequency band as the RX.One exception is [13], where the RX was measured while aco-integrated TX was simultaneously operating. However, theTX in [13] was only generating a weak output power around−18 dBm during the measurement.

In this work, we report a single-chip in-band full-duplextransceiver for EPR spectroscopy, where TX power is tar-geted at >20 dBm while RX is operating at the same timeand frequency band. In this design, the TX self-interferenceproblem is mitigated by an on-chip cancellation circuit. Thisimproves RX performance through canceling the TX self-interference by at least 32 dB. For the first time, operationof a self-interference-resilient RX is demonstrated while aco-integrated TX is operating and generating >20 dBm outputpower. Moreover, low NF at low baseband frequencies isachieved. When the TX and self-interference cancellationcircuits are turned off, the RX achieves a measured NFof 3.1 dB/6.3 dB at 10 MHz/50 kHz baseband frequency,with a 60 kHz flicker noise corner. This flicker noise corneris more than 10× lower than the prior publications [15], [16].At −10 dBm self-interference power, the measured RX NFis 6.8 dB/11.1 dB at 10 MHz/50 kHz baseband frequencywhen the self-interference cancellation circuit is turned on. TheNF increases by 5.6 dB/9.6 dB at 10 MHz/50 kHz basebandfrequency when the self-interference cancellation circuit isturned off, demonstrating the functionality and importance ofthe self-interference cancellation circuit. Utilizing this trans-ceiver, an EPR spectrometer is assembled and characterized bymeasuring a 2,2-Diphenyl-1-Picrylhydrazyl (DPPH) powdersample. At −10 dBm self-interference power at the input ofthe RX, the SNR of the EPR signal improves by 7 dB whenthe cancellation circuit is turned on. Compared to the previouswork presented in [11], the reported EPR spectrometer showsa SNR improvement of 25 dB.

The paper is outlined as follows. Section II discusses thebackground of EPR spectrometers and the importance of theself-interference cancellation. Section III describes the designof the single-chip transceiver with self-interference cancella-tion circuit. Section IV reports the measurement results ofthe transceiver, as well as the assembled spectrometer, andSection V concludes the paper.

II. BACKGROUND

In this section, the background of EPR spectroscopy is firstintroduced. Next, the importance of self-interference cancella-tion is discussed.

Fig. 1. (a) Energy absorption of an EPR sample versus DC magnetic field.(b) Architecture of a typical continuous-wave absorption-power-based EPRspectrometer.

A. EPR and EPR Spectrometer

From the quantum physics theory, it is known that electronswith spin quantum number Ms possess a magnetic momentof gβMs. Here, g is a material-dependent constant, andβ is equal to 9.2740154 × 10−24 JT−1. When an electronis placed inside a magnetic field B0, it obtains an energyof U = B0gβMs. For a spin 1/2 system where possible valuesfor Ms are + 1/2 and −1/2, there exist two possible electronspin energy levels, and the energy difference between thetwo energy levels is equal to �U = B0gβ. In materialswhere empty electron spin energy levels exist, an electrontransition from the lower energy level to the higher energylevel may be induced by an electromagnetic field B1 withappropriate frequency f , where the photon energy satisfiesh f = �U = B0gβ. The frequency f typically falls intothe microwave regime. In reality, due to effects such asthe interaction between electrons with nearby electrons andlattice, for a given microwave frequency f , the absorption ofenergy may occur across a range of magnetic field centeredaround B0, as shown in Fig. 1(a). This phenomenon, thatcertain materials, when placed inside a magnetic field, willabsorb incoming microwave energy at a particular frequency,is called Electron Paramagnetic Resonance (EPR).

In order to measure the absorption curve of a sampleshown in Fig. 1(a), typically an EPR spectrometer is used.In this work, we focus on continuous-wave absorption-power-based EPR spectrometer, as it is the most widely-usedEPR spectrometer. Fig. 1(b) shows the architecture of anabsorption-power-based EPR spectrometer. During the mea-surement, the TX delivers a microwave signal to the resonator,which contains the sample. The resonator generates a strong

2410 IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 51, NO. 10, OCTOBER 2016

electromagnetic field B1, which interacts with the sample andexcites electron transition. The amplitude of the magneticfield B0 is gradually swept, but the frequency and power of theTX signal is fixed. The RX receives the reflected power fromthe resonator, which varies when the sample absorbs part of themicrowave signal inside the resonator. This absorption dependson the DC magnetic field inside the resonator and reaches amaximum value as shown in Fig. 1(a). Since this absorptionis small, it is difficult to measure the change in the reflectedsignal coupled to the receiver. This is because the changeresults in a small DC signal after down-conversion. To increasethe sensitivity of the detection and reduce the effects of low-frequency noise (1/ f ), a small amplitude-modulated magneticfield (e.g., 0.1 Gauss) is applied on top of a large DC magneticfield by injecting a low-frequency AC current (e.g., 100 kHz)to the magnet coil. This technique shifts the baseband signalfrom DC to the modulation frequency (e.g., 100 kHz). In addi-tion, the modulation mechanism converts the EPR curve, theamplitude of the base-band signal versus the DC field B0, to afirst-order derivative of the absorption line shown in Fig. 1(a).Theoretically, to maximize the sensitivity of the system, themodulation frequency should be much higher than the flickernoise corner of the RX. Unfortunately, due to limitationssuch as the difficulty of generating a strong high-frequencymagnetic field, the modulation frequency in a typical EPRspectrometer rarely exceeds 100 kHz. This poses a stringentrequirement on the RX NF at low baseband frequencies.

B. Background on TX Self-Interference

In an absorption-power-based EPR spectrometer, theminimally-detectable number of electron spins is equal to

(4kB

g2β2μ0VsT

) (�ω

ω0

) (1

Qη

) √FkB T

P(1)

where Vs and T are the volume and temperature of the sample,respectively. kB is the Boltzmann constant. μ0 is the vacuumpermeability. �ω is the spectral line width of the sample,which is a material-dependent constant. η and Q are the fillingfactor and quality factor of the resonator, respectively. ω0 isthe operational frequency of the EPR spectrometer. F is thenoise factor of the RX, and P is the TX output power [12].The equation is derived by comparing the EPR signal power ofthe sample and the noise power of the RX, and the minimallydetectable number of electron spins is defined when these twopowers are equal (SNR = 1). Clearly, the RX NF shouldbe reduced and the TX output power should be maximized.Note that the equation above has not included the samplesaturation effect, where the EPR signal power of the samplemay stop to increase if the TX power exceeds the samplesaturation power. Existing studies have shown that the samplesaturation power is strongly sample-dependent and can be aslarge as several tens of milliwatts [18]. Therefore, in thiswork, the maximum TX output power is targeted at around20 dBm. Unfortunately, for this amount of TX output power,it is difficult to achieve low RX NF using prior single-chip-transceiver-based EPR spectrometers [11]. This is caused bythe TX self-interference signal present at the RX input. During

the operation of an EPR spectrometer, a small portion of theTX signal, denoted as TX self-interference, always couplesto the RX input through two major coupling mechanisms:1) the finite isolation value of the circulator, and 2) thereflection from the resonator due to non-perfect matching. As acommercial circulator typically does not provide an isolationof better than 30 dB, the TX self-interference signal is about30 dB lower than the TX output power. The frequency of theself-interference signal only deviates from the RX frequencyby the modulation frequency of the B0 field, which is nomore than 100 kHz. Because the frequency of the TX self-interference and EPR signals deviate by such a small amount,it is extremely challenging to remove the TX self-interferencesignal before down-conversion by a filter. As indicated before,the baseband frequency for an EPR spectrometer is same asthe modulation frequency of the magnetic field B0. In suchlow frequencies, the flicker noise of mixers and basebandamplifiers can contribute significantly to the RX NF, if it is notsuppressed by a strong amplification before down-conversion.However, if the TX is generating a strong output signal, the TXself-interference would be excessively large at the RX input,saturating the RX and preventing a large signal gain. As aresult, a large TX output power and a low RX NF cannot besimultaneously achieved without using a novel architecture.For example, the transceiver in [11] achieved a good RX NFat the cost of a relatively low TX output power of around5 dBm. This limitation significantly impacted the sensitivityof the spectrometer.

In recent years, several techniques have been proposedthat could mitigate the TX self-interference in an in-bandfull-duplex transceiver [13]–[17] that targeted at wirelesscommunication applications. However, as described in theintroduction section, these works provide poor NF at lowbaseband frequencies (for example, the flicker noise cornerwas 2 MHz in [15]), and thus they are not suitable for EPRspectroscopy.

III. ARCHITECTURE OF THE TRANSCEIVER WITH TXSELF-INTERFERENCE CANCELLATION

In this work, in order to improve the sensitivity of theEPR spectrometer, we mitigate the TX self-interference byintegrating a cancellation circuit on the transceiver chip. In thissection, we first introduce the architecture of the transceiver.Next, the details of the self-interference cancellation circuit,TX, and RX are discussed.

A. Transceiver Architecture

As shown in (1), the sensitivity of an EPR spectrometerincreases with the operation frequency. Therefore, it is attrac-tive if the frequency can be maximized. However, in this work,a planar PCB-based resonator is used to minimize the costand form factor of the spectrometer, and make it suitablefor planar applications. Therefore, due to the challenges offabricating a large planar high-Q resonator at high frequencies,the frequency of the prototype transceiver is designed to bearound 4.5 GHz.

YANG AND BABAKHANI: FULL-DUPLEX SINGLE-CHIP TRANSCEIVER WITH SELF-INTERFERENCE CANCELLATION IN 0.13 μm SiGe BiCMOS 2411

Fig. 2. Architecture of the reported transceiver.

The architecture of the proposed transceiver is presentedin Fig. 2. In order to cancel the TX self-interference signal,an on-chip self-interference cancellation circuit is insertedbetween the TX and RX. The cancellation circuit takes the TXoutput as its input and then generates a cancellation signal withthe same amplitude but a 180 phase shift, relative to the TXself-interference signal at the input of the receiver. In orderto achieve a low RX NF, the noise penalty from the self-interference cancellation circuit must be minimized. Towardthis purpose, the cancellation signal is injected after the low-noise amplifier (LNA), instead of at the RX input. By carefullychoosing the gain of the LNA, the noise contribution fromthe cancellation circuit can be suppressed, while the RX stillmaintains good linearity.

After the TX self-interference signal is removed from theRX, the desired EPR signal is further amplified and then down-converted to baseband. At the baseband input, a DC offsetfiltering block is used to reject the DC offset. Because thebaseband signal in EPR spectroscopy is typically on the orderof tens of kHz, external 10μF capacitors are utilized in theDC offset filtering block. They can potentially be removedby implementing on-chip DC offset cancellation circuitriessuch as the one reported in [19]. After filtering, the remainingEPR signal is further amplified by baseband amplifiers anddelivered to an off-chip data-acquisition unit for processing.

B. Self-Interference Cancellation Circuit

The self-interference cancellation circuit must generate acancellation signal with tunable phase and amplitude to cancelthe interference signal. In this work, no assumption is made onthe phase of the interference signal, so the cancellation circuitis required to have 360° phase tunability. The amplitude of theTX self-interference signal is assumed to have a maximumvalue of approximately −10 dBm, considering 20 dBm TX

output power and 30 dB TX-RX isolation. As introducedbefore, the cancellation signal is injected into RX after theEPR signal has been amplified by a LNA, whose gain isdesigned to be approximately 10 dB in this work; therefore,the maximum amplitude of the cancellation signal is requiredto exceed 0 dBm. Moreover, as the RX input-referred P1dB isestimated to be −25 dBm, it is desirable that the cancellationsignal can have at least 15 dB of amplitude tunability so that aself-interference signal as high as −10 dBm can be cancelledas well. Fortunately, since the self-interference is a single-tonesignal, the linearity requirement for the cancellation circuit isrelatively relaxed, as no intermodulation signal with frequencyclose to the EPR signal will be generated.

Fig. 3 presents the schematic of the self-interference cancel-lation circuit. Because the TX output exceeds 20 dBm, it mustbe attenuated before being sent to the input of the cancellationcircuit; otherwise, the active devices in the cancellation circuitmay face severe reliability/breakdown concerns. As shownin Fig. 3, the attenuator at the input of the cancellation circuitis realized through a voltage divider. In order to reduce thenoise from the self-interference cancellation circuit, we havemerged the amplitude tuner with the attenuator instead ofadding an extra stage to perform amplitude tuning for the can-cellation signal. In this work, instead of a resistor divider, tworeverse-biased diode varactors are used to perform the ampli-tude tuning to minimize the added thermal noise. By changingthe capacitance of the varactors, the dividing ratio is varied.In a divider containing only two capacitors C1 and C2, asshown in Fig. 4 (a), if the varactor capacitance C1 can beincreased by N×, without considering the load impedance,the amplitude tuning range of the divider output is

NC1 + C2

C1 + C2(2)

2412 IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 51, NO. 10, OCTOBER 2016

Fig. 3. Schematic of the self-interference cancellation circuit.

Fig. 4. (a) A simple capacitor divider. (b) Inductor-assisted capacitor divider with an ideal inductor. (c) Inductor-assisted capacitor divider with a real inductor.

To achieve a large amplitude tuning range at the divideroutput, the tuning range of the varactor capacitance should bemaximized. Unfortunately, in the process where the proposedtransceiver is implemented, the tuning range of the varactorcapacitance is limited to approximately 2–3×. Therefore, theamplitude tunability of the divider output is restricted to 10 dBor less. This number could be further reduced if the loadimpedance is considered. In order to enlarge the amplitudetuning range of the divider output, in this work, a parallel LCresonance network is used to increase the tuning range for theeffective capacitance of the varactor. As shown in Fig. 4(b),if an inductor is connected in parallel to the varactor, theadmittance of the network becomes

Y = jωC1 + 1

jωL(3)

If the value of L and C1 are chosen such that the admittancealways has a positive imaginary part, the effective capacitanceof this LC network would be

Cef f = C1 − 1

ω2 L(4)

Now the tuning range of Ceff is

NC1 − 1ω2 L

C1 − 1ω2 L

(5)

By choosing C1 slightly larger than 1ω2 L

, the tuning rangeof Ceff could be much larger than 3, and thus the amplitudeof the divider output could achieve a large tuning range.Moreover, because the admittance of the LC network is purely

imaginary and remains positive, the phase shift of the voltagedivider is constant across all the dividing ratios.

In reality, the quality factor of the inductor may limit theamplitude tuning range of the divider output. If the inductorhas a quality factor of Q, the admittance of the parallel LCnetwork becomes

Y = jωC1 + 1

jωL + ωLQ

= jωC1 − jωL

ω2 L2(

1 + 1Q2

) +ωLQ

ω2 L2(

1 + 1Q2

)

≈ jωC1 − j

ωL+ 1

ωL Q(6)

Note that the effective capacitance remains the same, as in thecase where the inductor is ideal, but the admittance now hasa positive real part. The resulting tuning range of the dividingratio is

1 + jω2L Q(C2 + Cef f −max)

1 + jω2L Q(C2 + Ce f f −min)(7)

Here Ceff-max and Ceff-min are the maximum and minimumvalues of Ceff. This equation shows that, for the same valuesof Ceff and C2, a lower inductor Q results in a lower tuningrange of the dividing ratio. Moreover, the phase shift of thedivider output would vary at different dividing ratios due tothe finite inductor Q. Fig. 5 plots the simulated output of thevoltage divider under different inductor Q. The simulation isperformed at 4.5 GHz. C2 is set at 200 fF, the inductanceis chosen to be 0.8 nH, and C1 is varied from 2 pF to 5 pF.As shown in the figure, the inductor Q should be maximized to

YANG AND BABAKHANI: FULL-DUPLEX SINGLE-CHIP TRANSCEIVER WITH SELF-INTERFERENCE CANCELLATION IN 0.13 μm SiGe BiCMOS 2413

Fig. 5. (a) Simulation results for the dividing ratio under different inductor Q. (b) Simulation results for the phase shift of the divider output under differentinductor Q.

achieve a large tuning range for the divider ratio and maintaina small variation in the phase shift of the divider output. In thiswork, with an inductor Q of 13, the divider output has a tuningrange of 14 dB, and the phase shift across the entire tuningrange is within 4 in the simulation.

In this work, a high input-impedance buffer follows thedivider. The input impedance of the buffer can be absorbedinto the quality factor of the inductor when analyzing theperformance of the voltage divider. Because the amplitude ofthe divider output can still be large, high breakdown transistorsare used when constructing the buffer. The load impedance ofthe buffer is designed to be variable so that 4 dB additionaltuning of the amplitude of the cancellation signal can beachieved. An I-Q generator follows the buffer and convertsthe differential signal into quadrature. The quadrature signalis sent to a Cartesian phase shifter, which assigns differentweights to the I and Q branches of the signal, and thencombines them. The weights of the I and Q signals arecontrolled by varying the gain of the two buffers that amplifythe signal. The cancellation is finally injected into the RXthrough a buffer.

One of the most important criteria of the self-interferencecancellation circuit is that its noise contribution must be small.To achieve this goal, voltage amplification in the cancellationcircuit is mostly avoided, as any voltage gain will amplifythe noise of the cancellation circuit as well. Fig. 6 presentsthe simulation results of the noise contribution from thecancellation circuit. The simulation is performed when the gainof the cancellation circuit is set at the maximum value. At theLNA output, where the cancellation signal is injected to theRX, the noise from the cancellation circuit is much smallerthan the noise caused by the LNA (including the noise fromthe 50� LNA input and the noise produced by the LNA). Thistranslates to a noise figure penalty of only 0.2 dB.

C. Design of the TX and RX

In this section, the design of the TX and RX of thetransceiver chip is presented.

In this work, the TX signal is generated using a voltage-controlled oscillator (VCO). The frequency of the VCO isdesigned to be twice the operation frequency of the EPR spec-trometer. A frequency divider then divides the frequency of the

Fig. 6. Simulated noise power at the LNA output and the cancellationcircuit output when the cancellation circuit gain is set at the maximum value.The self-interference cancellation circuit introduces a simulated noise figurepenalty of only 0.2 dB.

VCO signal by two, simultaneously providing I and Q signalsthat are necessary to generate the LO signals for the down-conversion mixer. Fig. 7 presents the schematic of the VCOand the frequency divider. The VCO takes a differential cross-coupled design. As the VCO frequency is around 9 GHz, thefrequency divider uses a current-mode logic (CML) structure,which is suitable for high-speed operation.

Following the frequency divider, an amplifier chain ampli-fies the TX signal. A power amplifier (PA) finally delivers thesignal to a 50� off-chip load through an external balun. Theschematic of the PA is shown in Fig. 8. The PA is based ona class-AB design and is fully integrated on the same chip.The simulated single-ended output-referred 1 dB compressionpoint is 20 dBm.

The design of the RX front-end is presented in Fig. 9. Thefirst stage in the RX is an inductive-degenerated cascode LNA.The load of the LNA is shared with the buffer of the self-interference cancellation circuit, which injects the cancellationsignal at the LNA output. In order to monitor the effectsof self-interference cancellation, a voltage-monitoring bufferis added at the LNA output. This buffer senses the LNA

2414 IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 51, NO. 10, OCTOBER 2016

Fig. 7. Schematic of the VCO and the CML frequency divider.

Fig. 8. Schematic of the PA.

output, attenuates it by 10 dB, and then sends it to outputpads. The output of the monitoring buffer is used to guide theamplitude and phase tuning of the cancellation signal. Afterthe TX self-interference signal has been removed from the RX,the remaining EPR signal is further amplified to suppress theflicker noise caused by mixers and baseband amplifiers. In thiswork, the down-conversion is performed by a current-modepassive mixer. Compared to active mixers, the passive mixerintroduces a smaller flicker noise. Moreover, by minimizingthe input impedance of the baseband trans-impedance ampli-fier (TIA), a current-mode passive mixer provides an improvedlinearity. This is because the large voltage swing, which causesdevice nonlinearities, is avoided in the circuit. The LO signalof the passive mixer is generated from the I/Q outputs of thefrequency divider following the VCO. The signal has a 25%duty cycle, increasing the mixer conversion gain by 3 dB,compared to a 50% duty cycle LO [20]. The design of the Gmcell is also shown in Fig. 9. This block is constructed using a

class-AB CMOS cell. As discussed in [21], such Gm cell hasvery high linearity. In simulation, the trans-conductance gainis only reduced by 1 dB when the single-ended input voltageswing reaches 1.4 V.

The design of the baseband amplifiers is shown in Fig. 10.The baseband TIA is implemented using a common-baseamplifier. The input impedance is kept small, at 10�, toincrease the linearity of the Gm cell and the passive mixer.The signal is finally amplified and delivered off-chip througha 50� buffer.

IV. MEASUREMENT RESULTS

In this section, we first report the measurement results ofthe integrated transceiver, and then present the measured EPRresponse of various samples.

A. Transceiver Measurement Results

Fig. 11 shows the micrograph of the reported transceiver.The transceiver chip is fabricated in IBM 0.13 μm BiCMOSprocess technology. It occupies an area of 4.8 mm× 2.5 mmand consumes 2 W of DC power during the operation. Majorblocks of the transceiver are labeled in the figure.

The measured frequency tunability of the TX signal isshown in Fig. 12. During this measurement, the voltage of thetuning node of the on-chip VCO is varied, and the frequencyof the frequency divider output is measured using an Agilentsignal analyzer N9030A. As shown in the figure, the frequencyof the signal can be varied from 3.8 GHz to 5.2 GHz.

The TX output power versus frequency is shown in Fig. 13.This power is measured after the differential TX outputs havebeen combined using an off-chip balun. It is observed that theTX can generate a maximum output power of 22 dBm. Themeasured 3 dB bandwidth is 1 GHz (from 3.8 to 4.8 GHz.)

The performance of the self-interference cancellation circuitis shown in Fig. 14. The purpose of this measurement is to

YANG AND BABAKHANI: FULL-DUPLEX SINGLE-CHIP TRANSCEIVER WITH SELF-INTERFERENCE CANCELLATION IN 0.13 μm SiGe BiCMOS 2415

Fig. 9. Schematic of the RX front-end.

Fig. 10. Schematic of the baseband amplifiers.

Fig. 11. Microphotograph of the fabricated chip.

determine if the cancellation circuit can cancel the TX self-interference signal across a wide range of phases, amplitudes,and frequencies. During this measurement, the TX output isconnected to the RX input through an external phase shifterand a variable-gain attenuator. Therefore, the phase and ampli-tude of the self-interference signal can be externally controlledand varied. Moreover, the frequency of the self-interferencesignal can be controlled by varying the oscillation frequency ofthe on-chip VCO. As the phases, amplitudes, and frequency of

Fig. 12. Measurement results for the VCO tuning range.

the TX self-interference signal are varied, the control voltageof the on-chip self-interference cancellation circuit is changedto cancel the self-interference signal. As a proof-of-conceptdesign, in this work, the control voltages are manually tunedusing an external voltage supply with a step voltage of 0.01 V.In future works, a microcontroller can be used to automaticallysearch for and generate the optimal tuning settings for thecancellation circuit. The measurement results are presentedin Fig. 14. Clearly, the TX self-interference signal can becancelled by at least 32 dB across a wide range of phases,amplitudes, and frequencies.

The RX conversion gain and linearity are studied nextin Fig. 15. As shown in this figure, the RX achieves aconversion gain of 46 dB. If the self-interference cancel-lation circuit is turned off, the self-interference P1dB is−20 dBm. However, when the self-interference cancellationcircuit is turned on and the TX self-interference signal iscancelled, the self-interference P1dB increases to −8 dBm.

2416 IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 51, NO. 10, OCTOBER 2016

Fig. 13. Measured TX output power versus frequency.

Fig. 14. (a) Measured self-interference cancellation under different inter-ference power and phase. (b) Measured self-interference cancellation underdifferent frequencies.

This demonstrates the effectiveness of the self-interferencecancellation circuit.

Finally, the NF of the RX is studied. During the NFmeasurement, a stable external RF source is used to generate

Fig. 15. Measurement results for the RX conversion gain.

Fig. 16. Measurement results for the RX NF under various conditions.

the TX and LO signals for the down-conversion mixer. Thisimproves measurement accuracy, especially at low basebandfrequencies. As shown in Fig. 16, when both the TX and thecancellation circuit are turned off, the RX achieves a NF of3.1 dB/6.3 dB at 10 MHz/50 kHz baseband frequencies. Whenthe TX remains off and the cancellation circuit is turned onand configured at the maximum gain settings, the cancellationcircuit incurs a noise penalty of no more than 0.6 dB acrossthe baseband frequencies from 50 kHz to 10 MHz. Whenthe TX and cancellation circuit are turned on and the self-interference power is −10 dBm at the RX input, the RX NFis 6.8 dB/11.1 dB at 10 MHz/50 kHz baseband. We believethat the degradation in the NF can be attributed to multiplefactors, such as the increased noise of the power supply andground rail caused by the TX. If the cancellation circuit isturned off under the same self-interference power, the RX NFis 12.4 dB/20.7 dB at 10 MHz/50 kHz baseband frequencies.This demonstrates that the self-interference cancellation circuitcan improve the NF significantly when there is a strong TXself-interference signal at the RX input.

YANG AND BABAKHANI: FULL-DUPLEX SINGLE-CHIP TRANSCEIVER WITH SELF-INTERFERENCE CANCELLATION IN 0.13 μm SiGe BiCMOS 2417

TABLE I

PERFORMANCE SUMMARY AND COMPARISON WITH PRIOR ARTS

Table I compares the reported transceiver with prior in-bandfull-duplex transceivers. The reported transceiver achieves thelowest NF, especially at low baseband frequencies. The flickernoise corner of the reported transceiver is 60 kHz, more thanone order of magnitude lower than prior works. Furthermore,

the reported transceiver is the first work that demonstrates theoperation of the in-band full-duplex RX and the associatedself-interference cancellation circuit when a co-integrated TXis operating at the same time and frequency band and gener-ating > 20 dBm power.

2418 IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 51, NO. 10, OCTOBER 2016

Fig. 17. Picture of the spectrometer setup.

Fig. 18. Measured EPR response from DPPH powder sample in (a) log and(b) linear scale.

B. EPR Spectrometer Measurement Results

Utilizing the reported transceiver, an EPR spectrometer isbuilt and tested, the picture of which is shown in Fig. 17.The resonator of the spectrometer has a quality factor ofapproximately 30. Its design details can be found in our earlierpublication [11]. Fig. 18 plots the measured EPR response ofa DPPH sample, using the assembled spectrometer. Duringthe measurement, the frequency of the TX signal is fixed, and

the B0 field is swept with a 100 kHz modulation. The TXdelivers around 20 dBm output power, and the self-interferencepower at the input of the RX is approximately −10 dBm.As shown in Fig. 18, if the cancellation circuit is configuredto cancel the TX self-interference, the resulted SNR increasesby approximately 7 dB. This demonstrates the effectivenessof the self-interference cancellation circuit. Compared to thepreviously reported EPR spectrometer based on a single-chiptransceiver, the reported spectrometer shows 25 dB higherSNR by measuring the same sample [11].

V. CONCLUSION

In this paper, we report an in-band full-duplex transceiverfor EPR spectroscopy application. The transceiver utilizes aself-interference cancellation circuit to mitigate the TX self-interference problem. In this work, the TX can deliver 20 dBmpower while maintaining a low NF of RX under a strongTX self-interference signal. The transceiver has the lowestNF and flicker noise corner, compared to prior in-band full-duplex transceivers. Using the reported transceiver, an EPRspectrometer is built and tested. Compared to previous EPRspectrometers based on single-chip transceivers, the reportedEPR spectrometer improves the SNR by 25 dB.

REFERENCES

[1] J. A. Weil and J. R. Bolton, Electron Paramagnetic Resonance: Elemen-tary Theory and Practical Applications. New York, NY, USA: Wiley,2006.

[2] S. S. Velan, R. G. S. Spencer, J. L. Zweier, and P. Kuppusamy, “Electronparamagnetic resonance oxygen mapping (EPROM): Direct visualizationof oxygen concentration in tissue,” Magn. Reson. Med., vol. 43, no. 6,pp. 804–809, Jun. 2000.

[3] J. Jiang, T. Nakashima, K. J. Liu, F. Goda, T. Shima, and H. M. Swartz,“Measurement of PO2 in liver using EPR oximetry,” J. Appl. Physiol.,vol. 80, no. 2, pp. 552–558, 1996.

[4] J. Matsumoto et al., “Absolute oxygen tension (pO2) in murine fattyand muscle tissue as determined by EPR,” Magn. Reson. Med., vol. 54,no. 6, pp. 1530–1535, Dec. 2005.

[5] L. Montanari, M. Clericuzio, G. Del Piero, and R. Scotti, “Asphalteneradicals and their interaction with molecular oxygen: An EPR probe oftheir molecular characteristics and tendency to aggregate,” Appl. Magn.Reson., vol. 14, no. 1, pp. 81–100, Feb. 1998.

[6] M. A. Volodin, G. V. Mamin, V. V. Izotov, and S. B. Orlinskii, “High-frequency EPR study of crude oils,” J. Phys., Conf. Ser., vol. 478,pp. 1–4, 2013.

YANG AND BABAKHANI: FULL-DUPLEX SINGLE-CHIP TRANSCEIVER WITH SELF-INTERFERENCE CANCELLATION IN 0.13 μm SiGe BiCMOS 2419

[7] Y.-H. Lee and J. W. Corbett, “EPR studies of defects in electron-irradiated silicon: A triplet state of vacancy-oxygen complexes,” Phys.Rev. B, vol. 13, pp. 2653–2666, Mar. 1976.

[8] J. Isoya, T. Umeda, N. Mizuochi, N. T. Son, E. Janzén, and T. Ohshima,“EPR identification of intrinsic defects in SiC,” Phys. Status Solidi B,vol. 245, no. 7, pp. 1298–1314, Jul. 2008.

[9] J. Anders, A. Angerhofer, and G. Boero, “K-band single-chip elec-tron spin resonance detector,” J. Magn. Reson., vol. 217, pp. 19–26,Aug. 2012.

[10] T. Yalcin and G. Boero, “Single-chip detector for electron spin resonancespectroscopy,” Rev. Sci. Instrum., vol. 79, no. 9, p. 094105, 2008.

[11] X. Yang and A. Babakhani, “A single-chip electron paramagneticresonance transceiver in 0.13-μm SiGe BiCMOS,” IEEE Trans. Microw.Theory Techn., vol. 63, no. 11, pp. 3727–3735, Nov. 2015.

[12] C. P. Poole, Jr., and H. A. Farach, Eds., Handbook of Electron SpinResonance, vol. 2. New York, NY, USA: Springer, 1999.

[13] D. Yang and A. Molnar, “A widely tunable active duplexing transceiverwith same-channel concurrent RX/TX and 30 dB RX/TX isolation,” inProc. IEEE Radio Freq. Integr. Circuits Symp., Jun. 2014, pp. 321–324.

[14] J. Zhou, T.-H. Chuang, T. Dinc, and H. Krishnaswamy, “Receiver with>20 MHz bandwidth self-interference cancellation suitable for FDD,co-existence and full-duplex applications,” in IEEE ISSCC Dig. Tech.Papers, Feb. 2015, pp. 342–343.

[15] D.-J. van den Broek, E. A. M. Klumperink, and B. Nauta, “A self-interference-cancelling receiver for in-band full-duplex wireless with lowdistortion under cancellation of strong TX leakage,” in IEEE ISSCC Dig.Tech. Papers, Feb. 2015, pp. 344–345.

[16] D.-J. van den Broek, E. A. M. Klumperink, and B. Nauta, “A self-interference cancelling front-end for in-band full-duplex wireless andits phase noise performance,” in Proc. IEEE Radio Freq. Integr. CircuitsSymp., May 2015, pp. 75–78.

[17] J. Zhou, N. Reiskarimian, and H. Krishnaswamy, “Receiver with inte-grated magnetic-free N-path-filter-based non-reciprocal circulator andbaseband self-interference cancellation for full-duplex wireless,” in IEEEISSCC Dig. Tech. Papers, Jan./Feb. 2016, pp. 178–180.

[18] B. Sanyal, S. P. Chawla, and A. Sharma, “An improved method toidentify irradiated rice by EPR spectroscopy and thermoluminescencemeasurements,” Food Chem., vol. 116, no. 2, pp. 526–534, Sep. 2009.

[19] S. Zhou and M. C. F. Chang, “A CMOS passive mixer with low flickernoise for low-power direct-conversion receiver,” IEEE J. Solid-StateCircuits, vol. 40, no. 5, pp. 1084–1093, May 2005.

[20] A. Mirzaei, H. Darabi, J. C. Leete, and Y. Chang, “Analysis andoptimization of direct-conversion receivers with 25% duty-cycle current-driven passive mixers,” IEEE Trans. Circuits Syst. I, Reg. Papers, vol. 57,no. 9, pp. 2353–2366, Sep. 2010.

[21] D. Murphy et al., “A blocker-tolerant, noise-cancelling receiver suit-able for wideband wireless applications,” IEEE J. Solid-State Circuits,vol. 47, no. 12, pp. 2943–2963, Dec. 2012.

[22] Bruker Corporation Model ELEXSYS-II E560. [Online]. Available:https://www.bruker.com

Xuebei Yang received the B.S. degree in microelec-tronics from Tsinghua University, Beijing, China,in 2008, and the M.S. and Ph.D. degrees in electricaland computer engineering from Rice University,Houston, TX, USA, in 2010 and 2016, respectively.

He is currently a Senior Analog IC Design Engi-neer at Oracle Corporation.

Dr. Yang was the recipient of the IEEE MicrowaveTheory and Techhiques Society (IEEE MTT-S)Graduate Fellowship for medical applications andthe Chinese Government National Award for Out-

standing Self-financed Students Abroad, both in 2015.

Aydin Babakhani received the B.S. degree inelectrical engineering from Sharif University ofTechnology, Tehran, Iran, in 2003, and the M.S.and Ph.D. degrees in electrical engineering fromthe California Institute of Technology (Caltech),Pasadena, CA, USA, in 2005 and 2008, respec-tively.

He was a postdoctoral scholar at Caltech in 2009and a research scientist at IBM T. J. WatsonResearch Center in 2010. He is currently an Assis-tant Professor in the Electrical and Computer Engi-

neering Department and the Director of the Rice Integrated Systems andCircuits Laboratory at Rice University, Houston, TX, USA.

Dr. Babakhani has won multiple best paper awards including the BestPaper Award in IEEE RWS Symposium in 2016, the Best Paper Awardin IEEE RWS Symposium in 2015, the Best Paper Award in IEEE IMSSymposium in 2014, and the 2nd Place Best Paper Award in IEEE IMSSymposium in 2016 and IEEE AP-S Symposium on Antennas and Propagationin 2016. He has more than 70 publications in peer-reviewed journals andconferences as well as 20 patents that are issued or pending. He receiveda prestigious NSF CAREER award in 2015, an innovation award fromNorthrop Grumman in 2014, and a DARPA Young Faculty Award (YFA)in 2012. He is a co-founder of MicroSilicon Inc. He received the CaltechElectrical Engineering Department’s Charles Wilts Best Ph.D. Thesis Prizefor his work on Near-Field Direct Antenna Modulation (NFDAM). From2006 to 2008 he was the Vice Chair of the IEEE Microwave Theory andTechniques Society Metro LA/SFV Joint Sections MTT-S Chapter 17.1.He was the recipient of the Microwave Graduate Fellowship in 2007, theGrand Prize in the Stanford-Berkeley-Caltech Innovators Challenge in 2006,Analog Devices Inc. Outstanding Student Designer Award in 2005, as well asCaltech Special Institute Fellowship and Atwood Fellowship in 2003. He wasalso the Gold Medal winner of the National Physics Competition in 1998, andthe Gold Medal winner of the 30th International Physics Olympiad in 1999,in Padova, Italy.